The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to ingest and process highly sensitive personal data, including private chat histories from WeChat and QQ, and location data (GPS) extracted from photo metadata. While the skill processes data locally, the collection of this data represents a significant privacy and exposure risk.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of external data.
Ingestion points: The skill reads and parses chat logs (txt, json, html), social media content, and text files via scripts in the tools/ directory.
Boundary markers: The generated SKILL.md uses markdown headers to separate sections but lacks explicit instructions to the agent to ignore commands embedded in the distilled memory data.
Capability inventory: The skill uses the Bash tool to execute local scripts and the Write/Edit tools to manage files and generate new skills.
Sanitization: The parsing scripts do not sanitize the extracted text to prevent it from being interpreted as instructions by the agent during the distillation or interaction phases.
[COMMAND_EXECUTION]: The skill executes local Python scripts using the Bash tool to parse data, analyze photos, and manage file versions.
[COMMAND_EXECUTION]: The skill dynamically generates and writes new SKILL.md files and directory structures at runtime based on the analysis of user-provided data and templates.

create-crush